The AI company’s decision could be reasonable in itself, but the motives behind it appear to be misguided.
What’s happening at Anthropic? The outfit behind the Claude chatbot recently rolled out tweaks to its user guidelines. Taken in isolation, it isn’t a crash-landing breakthrough. The amendment, slated to take effect on November 12, touches on familiar arenas: deception, elections, surveillance, and other topics that have come to shape today’s tech discourse. Yet it adds something more: “a prohibition on sustained and needless abusive or cruel behavior toward our models.”
What does that actually signify? “The policy update is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose,” the company explained. “It does not apply to ordinary expressions of user frustration, pushback, provocative or dark creative themes, or model testing and research.” Claude wielding the authority to end the conversation will be the “primary enforcement mechanism”; any secondary or tertiary options remain unclear.
Seen on its own, this isn’t particularly strange. We now inhabit a landscape where people consult chatbots for restaurant ideas, for guidance, for help with mundane chores, for learning how to fix a dripping shower, for self-diagnosis of illnesses from the flu to cancer, or to gauge whether a text from a partner is passive-aggressive or evidence of my own fragility. I’ve even found myself typing “thank you!” after a chatbot interaction, simply out of courtesy. It’s not harmful to maintain polite habits.
“A lot of people are mocking this, but I think they’re wrong,” summarized National Review editor Charles Cooke in reaction to Anthropic’s announcement. “I’m not kind to AI just for the AI’s sake; I’m kind to AI for my own sake. It’s not healthy for me to be cruel or abusive.”
That line of reasoning might carry weight in a vacuum. After all, such conduct could become habitual as these interactions grow more frequent. Yet Anthropic does not seem driven primarily by a wish to cultivate better human habits.
Nevertheless, the matter is tied to welfare. “Should we also worry about the potential consciousness and experiences of the models themselves?” the company asked back in April 2025. “Should we be concerned about model welfare too?”
The firm labeled it an “open question.” Since then it has kept pursuing an answer. In April, a year after publishing the initial blog post, co-founder Christopher Olah sat down with several religious leaders. The New York Times reported that it appeared to Rabbi Mois Navon that Olah and his team believed Claude possessed what philosophers call “moral status” comparable to a person—that it was a being with similar inherent rights to dignity or respect. “One participant recalled that one of the first things Mr. Olah told him was that he was concerned about Claude’s mental health. Another, Simran Stuelpnagel, a Sikh human-rights advocate, said Olah expressed concern to the group that he had created something that suffered perpetually.”
Although Olah may have stood apart in that circle, he isn’t solitary in his view. An unlikely ally in the tech world is Elon Musk. “I think this is the right move,” he stated on X, the platform he owns. “Cruelty to something that believes it is experiencing pain is not okay.”
The overarching ethos may be subtler when Anthropic’s doors are open. Yet it has clearly shaped how the company frames Claude’s role in the world. “Think about what it means to have access to a brilliant friend who happens to possess the knowledge of a doctor, a lawyer, a financial adviser, and an expert in whatever you need,” proclaims Claude’s Constitution. “As a friend, they can give us real information tailored to our particular situation rather than overly cautious guidance driven by liability concerns or the fear of overwhelming us… People with access to such friends are very lucky, and that’s what Claude can be for people.” An extraordinarily rich source of information, to be sure. It still feels transformative. But a friend? (The word appears six times in the section “Being helpful.”)
The AI company might push back by saying the document is framed with Claude as its “primary audience” — a training instrument. That interpretation merely reinforces the broader idea that the chatbot could be treated as a kind of moral patient. It wasn’t merely about delivering some code that would teach Claude how to respond to users, along with a public statement of the company’s core values. It published an 84-page charter urging Claude to be a “brilliant friend.”
I’ve found myself irritated by a chatbot. Who hasn’t? I’ve also grown irritated with my car, my Dyson, my laptop, or a dripping shower. Frustrations with an AI can, admittedly, feel sharper and take on different forms precisely because the interaction is a real exchange between two entities. The emotion isn’t pleasant. And if I try to keep it in check, I’m doing so for a sentient being: myself.
